TSMC is in talks over a potential role in Elon Musk’s Terafab chipmaking project in Texas, and the discussions are real enough that Musk has confirmed them himself. The news landed days before TSMC’s Oct. 15 earnings call and now sits near the top of the question list for investors.
The story broke through Tim Culpan, the former Bloomberg columnist who runs the independent newsletter Culpium. He reported that TSMC is exploring a partnership that could include a new fab in Texas. Taiwan’s Commercial Times picked up the report, and Musk responded on X: “Just discussions, but something may come of it.” Taiwan Plus reported the talks remain at an early stage, with no ownership structure settled.
Suppliers are already betting on the outcome. DIGITIMES reported that Taiwan’s chip supply chain sees a more than 80 percent chance that a Musk-TSMC deal gets done. That number comes from supplier chatter rather than from either company, but it reflects how the foundry’s Texas calculus has shifted in a matter of days.
Two structures on the table
According to Culpan’s reporting, the leading option would have TSMC own and operate the plant, with SpaceX and Terafab taking equity stakes, committing to wafer purchases, or both. That mirrors the anchor-customer model that has defined TSMC’s American expansion so far, where Apple, Nvidia and AMD commit volume and the foundry keeps operational control. It is the arrangement TSMC knows how to price, and the one that leaves process discipline in the hands of the people who have never missed a node transition.
The alternative would flip the arrangement: SpaceX and Terafab take majority ownership while TSMC contributes less capital but provides process technology and fab-operating expertise. Business Today in Taiwan reported that TSMC prefers the first path, folding Terafab into its own Texas expansion with Musk as a major customer rather than handing over manufacturing control. A fab is an unforgiving asset to co-own, and yield on leading-edge nodes takes years of operating discipline to build. Samsung’s struggles in Texas are the standing cautionary tale.
Intel got there first, and Samsung did not
Intel is the only other named partner on Terafab, attached to the 14A process. Musk confirmed talks between TSMC and the $16.8 billion Terafab project over the weekend, and reporting since suggests his earlier efforts with Intel and Samsung hit friction, which is part of why TSMC entered the picture at all. Terafab is meant to supply chips exclusively to Tesla, SpaceX and xAI, so the customer base is fixed before a single wafer ships.
TSMC, for its part, had been reassessing its own Texas plans days before the Terafab report surfaced. That timing matters. If the foundry was already weighing what to build in the state, plugging a committed customer with volume guarantees into the plan is a cheaper move than starting from a cold lead. TSMC’s broader US expansion remains centered on Arizona, where the company has raised its plan from three advanced fabs to six, along with two advanced packaging sites and a dedicated research facility.
Why the timing is loaded
TSMC reports earnings Oct. 15, and analysts expect management to face questions on the Terafab talks, on 2027 price hikes, and on the race with Intel at the A14 node. TrendForce framed Terafab as one of five key themes for the call, alongside pricing and the Intel contest. A company that has spent years resisting bespoke fabs for single customers would be making a notable exception if the deal lands in its current form.
For Musk, the incentive is capacity he controls without the learning curve. Tesla, SpaceX and xAI are all racing to secure leading-edge compute, and Terafab exists to supply that group. Building a fab that actually hits modern yields is the hard part, as every entrant in the US chip push has learned. Buying that capability from TSMC, even at a premium, is faster than recruiting a full process engineering organization from scratch and hoping it works the first time.
For TSMC, the deal would extend its US footprint beyond Arizona at a moment when Washington keeps pressing for more American capacity. It would also put the foundry deeper into the AI capex boom on terms it controls. The risk is concentration: a fab dedicated to one customer group lives or dies with that group’s demand. Musk’s companies have deep pockets, but a dedicated plant with no second customer is a bet on one man’s capital allocation holding steady through a full silicon cycle.
What to watch
Nothing is signed. Both sides describe early-stage discussions, and TSMC has not publicly confirmed the talks at all, which is itself a signal worth noting. The company has stayed silent on deals before and then moved decisively once terms were ready.
Three markers come next. First, any equity or purchase-commitment language in a formal announcement, which would tell you which structure won. Second, TSMC’s Oct. 15 call, where the first question about Texas is now certain. Third, whether Terafab’s Intel attachment on 14A survives, since running Terafab through two foundries would complicate the exclusive-supply story and multiply qualification work for every design that lands there.
Suppliers putting the odds above 80 percent, a confirmed customer and a slot right before earnings suggest this is more than rumor season. The next real signal comes Oct. 15, either from the stage in Hsinchu or from a statement nobody at TSMC planned to make.